About the Role

We can train a great model. The next problem is making it fast enough to actually use in a real-time conversation — and that gap is enormous. A model that responds in 3 seconds is a demo. A model that responds in under 500ms is a product.

We’re looking for someone who specializes in taking trained models and squeezing every last millisecond out of them. You understand the full stack from model weights to serving infrastructure — quantization, KV cache optimization, kernel-level acceleration, batching strategies — and you know which lever to pull for which problem. You’ve worked with vLLM, SGLang, or similar frameworks at scale and have strong opinions about where they fall short.

This posting is aimed at experienced engineers and researchers who’ve operated at a senior to senior-staff level at big tech, a leading AI lab, or a high-traffic inference team. Everyone at Nuance is MTS — we don’t run title ladders — but we’re hiring people who have already done this work at scale.

Our stack is more complex than a standard LLM deployment: we’re serving a full-duplex multimodal system that must satisfy strict real-time latency constraints. There’s a lot of unsolved optimization work here, and we need someone who finds that genuinely exciting.

What You’ll Do

• Own end-to-end inference optimization across our model stack — LLMs, audio models, and diffusion-based components
• Implement and tune KV cache strategies for long-context conversations, including eviction policies, compression, and memory-efficient attention
• Evaluate, deploy, and extend inference serving frameworks (vLLM, SGLang, TensorRT-LLM, etc.) for our specific workloads
• Profile and benchmark end-to-end latency and throughput; identify and systematically eliminate bottlenecks
• Build internal tooling that makes optimization work faster and more rigorous — profiling viewers, end-to-end inference test harnesses, and other infrastructure that helps the team move quickly
• Accelerate diffusion model inference — consistency models, step distillation, caching strategies, and custom kernel optimizations
• Apply and develop quantization techniques (INT8, INT4, GPTQ, AWQ, and beyond) to reduce memory footprint and increase throughput without meaningfully degrading quality
• Work closely with research and infrastructure to ensure new models ship with optimized serving from day one

What We’re Looking For

• Significant hands-on experience with LLM inference optimization — you’ve shipped work on KV caching, memory layout, attention kernels, or batching strategies in a production or high-traffic research context
• Proven proficiency with inference serving frameworks — vLLM, SGLang, TensorRT-LLM, or similar — including going well beyond default configurations and adapting them to non-standard workloads
• Experience optimizing diffusion model inference (latency reduction, step distillation, caching, or kernel-level work)
• Strong Python and PyTorch skills; comfort reading and writing CUDA or Triton kernels is a significant plus
• A systematic approach to profiling and optimization — you measure first, then optimize
• Familiarity with speculative decoding or other inference-time acceleration techniques

Bonus Points

• Hands-on experience with post-training quantization (GPTQ, AWQ, or similar) and a clear sense of quality/performance tradeoffs
• Familiarity with multimodal or streaming inference architectures
• Experience deploying real-time AI systems with hard latency SLAs
• Prior work at an AI lab, inference startup, or on a high-traffic model serving platform
• Contributions to open-source inference frameworks

Compensation

$250,000 – $350,000 base salary, plus meaningful equity. We think long-term ownership matters and structure equity accordingly.

About the Role

Model quality is ultimately a data problem. The best architecture and the best training run can't outrun bad, slow, or poorly curated data — and at the scale we're operating, the difference between a good data pipeline and a great one shows up directly in the model.

We're looking for someone who lives and breathes data at scale. You know how to build pipelines that are fast, reliable, and maintainable — and you're just as comfortable taking a researcher's messy processing script and turning it into something that runs on petabytes as you are designing a new pipeline architecture from scratch. Research moves fast here, and the ability to productionize quickly without losing fidelity is the core skill.

Our data is multimodal — video, audio, and text — and the processing requirements are demanding: high throughput, low error rates, and strict quality filters. There's a lot of interesting engineering work here, and the impact is direct and measurable.

What You'll Do

• Design, build, and operate large-scale data pipelines for ingestion, processing, filtering, and curation of multimodal training data (video, audio, text)
• Take research-grade data processing code and turn it into robust, production-level pipelines — quickly and without losing correctness
• Optimize pipeline throughput and efficiency at scale; identify and eliminate bottlenecks across compute, I/O, and storage
• Build and maintain data quality systems — deduplication, filtering, validation, and quality scoring at scale
• Manage petabyte-scale datasets: storage architecture, versioning, lineage tracking, and cost efficiency
• Work closely with researchers to understand data requirements and translate them into scalable processing systems
• Build tooling and infrastructure that makes the research team faster — efficient data access, reproducible processing, and fast iteration loops

What We're Looking For

• Proven experience building and operating large-scale data pipelines in production — you've processed data at a scale where naive approaches break
• Strong proficiency with distributed data processing frameworks — Spark, Ray, Dask, or similar — and a clear sense of when to use each
• Solid software engineering fundamentals: you write clean, testable, maintainable code and understand why that matters when pipelines run unattended at scale
• Experience with multimodal data (video, audio) is a strong plus — understanding of formats, codecs, and processing libraries (FFmpeg, decord, etc.)
• Familiarity with ML data pipelines specifically — understanding of how data quality and format affect model training
• Ability to move fast: you can take a prototype script from a researcher and ship a production version in days, not weeks

Bonus Points

• Experience building data pipelines for large-scale model training (pre-training or fine-tuning)
• Familiarity with data versioning and lineage tools (DVC, Delta Lake, Apache Iceberg, etc.)
• Experience with streaming data pipelines or online data processing
• Prior work at an AI lab, video platform, or other data-intensive company
• Contributions to open-source data tooling

Compensation

$200,000 – $300,000 base salary, plus meaningful equity. We think long-term ownership matters and structure equity accordingly.

Logistics

• Location: In-person in Seattle, five days a week — we believe in the compounding value of working shoulder-to-shoulder.
• Visa sponsorship: We sponsor visas (O-1, H-1B, green card, etc.) from day one.
• AI-native tooling: Do your best work with the best tools, including unlimited tokens.

Benefits

• Health: We offer a variety of plans that meet your needs, including an HDHP with ~$2,000 in annual HSA contributions by the company (roughly 2x what most big tech companies put in).
• Time off: 15 days of PTO, 10 public holidays, and we close the office for a full week at year-end.
• Food: Lunch, drinks, and snacks on us every workday. We observe boba tea Tuesdays and Thursdays.
• Commuter benefits: Utilize pre-tax money (up to $340/month) for parking and transportation.
• 401(k): 4% match (100% of 1st 3% + 50% of next 2% contributions).

Nuance Labs is an equal opportunity employer. We believe diverse teams build better AI.